Enhancing Monoclonal Antibody Yield and Quality Through Automated Multi-Component Feedback Control Loops Using the MarqMetrix All-In-One Process Raman Analyzer

Significance of the topic

Real-time, multi-analyte process monitoring is a critical enabling technology for modern biomanufacturing. Raman spectroscopy provides specific molecular fingerprints in complex cell culture media, allowing non-destructive, in-line measurement of nutrients, metabolites, and product-related attributes. Deploying Raman as a Process Analytical Technology (PAT) enables closed-loop control strategies that directly couple cellular metabolic state to automated feeding decisions, improving yield, consistency, and critical quality attributes for monoclonal antibody (mAb) production.

Goals and overview of the study

This application study evaluated an advanced carbon-control feedback strategy implemented with the Thermo Scientific MarqMetrix All‑In‑One Process Raman Analyzer. The primary objective was to simultaneously monitor glucose and lactate in a 5 L fed‑batch CHO bioreactor and use those predictions to maintain a target total carbon concentration (glucose + lactate) of 2 g/L. The study compared three automated control logics: continuous glucose control at 8 g/L, automated fed‑batch bolus control (maintain 3–6 g/L), and the novel total carbon control. Outcomes assessed included titer, product quality (glycation), viable cell density, and viability.

Methodology

Chemometrics and data flow

• Partial least squares (PLS) models were developed for glucose and lactate using selected Raman spectral windows and preprocessing to maximize specificity and minimize overfitting.
• Glucose model used three spectral regions: 1065–1232 cm⁻¹, 1595–1863 cm⁻¹, and 2704–3078 cm⁻¹. Preprocessing: Savitzky–Golay first derivative (order 2, window 13), Standard Normal Variate (SNV), and mean centering.
• Lactate model used 800–1750 cm⁻¹ with Savitzky–Golay first derivative (order 2, window 11), L1 norm scaling (area normalization over 1540–1750 cm⁻¹), and mean centering.
• Model tuning used leave-one-out cross-validation and RMSECV to choose the number of latent variables, minimizing calibration and cross‑validation error while avoiding overfitting.

Bioreactor and culture conditions

• Cell line: CHO‑K1 GS in Efficient‑Pro medium with supplements (insulin 1.5 mg/L, 1% anti‑clumping), inoculation at 0.75 × 10⁶ cells/mL; run duration 14 days.
• Operational conditions: 37 °C, pH 7 ± 0.2 (controlled with CO₂ and sodium carbonate), dissolved oxygen 40%.
• Feeding: Standard fed‑batch feeding (days starting day 3) with Efficient‑Pro feed 3 and Enhancer, delivered either as bolus or continuous depending on the control mode.

Instrumentation and data integration

• Raman hardware and acquisition: Thermo Scientific MarqMetrix All‑In‑One Process Raman Analyzer with MarqMetrix Performance BallProbe Sampling Optic (also demonstrated with FlowCell optic). Acquisition used 785 nm laser, 450 mW, 3000 ms integration, averages = 20; data processed ~every 2 minutes to deliver near real‑time values.
• Data handling: Cosmic ray removal, averaging, timestamp alignment in an internal Python platform with downstream processing in SOLO 9.3.1 when applicable.
• Control loop: Raman predictions of glucose and lactate were sent to TruBio software and then to a DeltaV controller that calculated and dosed glucose via a pump to achieve the selected control logic (including total carbon target of 2 g/L).

Instrumentation used

• Thermo Scientific MarqMetrix All‑In‑One Process Raman Analyzer
• Thermo Scientific MarqMetrix Performance BallProbe Sampling Optic (and FlowCell option)
• TruBio control software (TruBio 6)
• DeltaV process control system and pump controller
• 5 L glass bioreactor (DynaDrive platform)

Main results and discussion

• Control performance and chemometrics: Raman‑based PLS models produced robust predictions across runs. Reported model statistics varied by dataset but included R² values roughly between 0.80 and 0.96 and standard error of prediction (SEP) values consistent with effective process control for both glucose and lactate.
• Process outcomes with total carbon control: Maintaining glucose + lactate at 2 g/L via Raman feedback increased final mAb titer by >10% relative to the other control modes.
• Product quality impact: Glycation (a deleterious modification linked to high glucose exposure) was strongly reduced under total carbon control. Measured glycation was ~2% in the total carbon run versus ~12% (continuous 8 g/L) and ~6% (fed‑batch bolus), corresponding to an 83% and 66% reduction relative to those two comparators, respectively.
• Cell health and viability: Viable cell density and viability were comparable or improved under the total carbon strategy versus continuous control; overall viability increased (reported improvements >15% in summary metrics) with total carbon management.
• Mechanistic rationale: By dynamically limiting glucose addition when lactate was high, the total carbon strategy promoted lactate consumption later in the run and prevented excessive lactate accumulation and prolonged high glucose exposure—conditions known to reduce cell health and promote glycation.

Benefits and practical applications

• Non‑destructive, simultaneous multi‑analyte measurement in a single scan simplifies process observability and reduces sampling burden.
• High update frequency and direct integration with process control systems enable advanced closed‑loop strategies (e.g., total carbon control) not feasible with off‑line analytics.
• Improvements in titer and critical quality attributes demonstrate direct commercial value: increased yield, better product quality (lower glycation), and improved batch-to-batch consistency.
• Flexible hardware (swappable probes, small footprint, stackable units) and chemometric model transferability support scaling and deployment across different cell lines and media.

Future trends and applications

• Extension to additional metabolites and CQAs: Application of multi-analyte Raman models to amino acids, ammonium, key glycosylation precursors, and product­-specific quality markers to expand closed‑loop control capabilities.
• Integration with AI and advanced control: Combining Raman PAT with model predictive control (MPC), reinforcement learning, or adaptive chemometrics to optimize feeding schedules and improve robustness in continuous and perfusion processes.
• Scale‑up and transferability: Further validation of model robustness across larger bioreactor scales and diverse manufacturing lines to support process standardization and regulatory readiness.
• Regulatory adoption and digitalization: Use of Raman PAT as a qualified measurement for real‑time release testing and automated quality assurance within digital manufacturing ecosystems.

Conclusion

This application study demonstrates that an in‑line, multi‑analyte Raman PAT platform can enable sophisticated feedback control strategies that materially improve monoclonal antibody production. Implementing a total carbon (glucose + lactate) control loop using MarqMetrix Raman predictions increased titer >10%, substantially reduced glycation, and improved viability versus conventional glucose control modes. The approach exemplifies how tight, multivariate monitoring combined with automated dosing and chemometric models creates a foundation for more consistent, high‑quality, and automated biomanufacturing.

References

1. Abu‑Absi NR, Kenty BM, Cuellar ME, et al. Real Time Monitoring of Multiple Parameters in Mammalian Cell Culture Bioreactors Using an In‑Line Raman Spectroscopy Probe. Biotechnology and Bioengineering. 2011;108(5):1215–1221.
2. Zhou M, Crawford Y, Ng D, et al. Decreasing Lactate Level and Increasing Antibody Production in Chinese Hamster Ovary Cells (CHO) by Reducing the Expression of Lactate Dehydrogenase and Pyruvate Dehydrogenase Kinases. Journal of Biotechnology. 2011;153(1):27–34.
3. Villa J, Khadka N, Keck K, Zhang L, Zustiak M. Process Raman as Platform Solution For Automated Glucose Feeding in Fed‑Batch Bioreactors. (Application note/internal report).
4. Villa J, Zustiak M, Ramirez D, et al. Demonstrating Chemometric Model Transferability for 5 Mammalian Cell Lines and 5 Media Types Using the Thermo Scientific MarqMetrix All‑In‑One Process Raman Analyzer to Monitor Upstream Bioprocesses. (Application note/internal report).
5. Villa J, Zustiak M, Kuntz D, Zhang L, Khadka N, Broadbelt K, Woods S. Use of Lykos and TruBio Software Programs for Automated Feedback Control to Monitor and Maintain Glucose Concentrations in Real Time. (Application note/internal report).